Generative Artificial Intelligence (“AI”) tools like ChatGPT, Scribe, Jasper, and others have catapulted exponentially in popularity in recent years, for widespread personal and professional uses supplementing, if not largely displacing, traditional search engines.
Applications for AI interactions in the workplace, algorithmically simulating human reasoning and inference, are expanding as quickly as users can draft new prompts requesting designs, how-to guides, correspondence, and countless other outputs. AI tools have quickly transitioned from an amusing new technology to essential tools for professionals and businesses, driving innovation and efficiency. These tools are used by businesses for an ever-expanding list of purposes, including brainstorming ideas based on patterns and data analysis; creating and memorializing documents, procedures, manuals, and tutorials; generating marketing and other client-facing materials; drafting communications; summarizing documents; explaining concepts and processes; and even generating code.
As these tools become more integrated into workplace processes, courts and litigants are beginning to confront the question of whether and to what extent AI searches and “chats” are discoverable in litigation. As the Federal Rules of Civil Procedure permit broad discovery regarding any nonprivileged matter that is relevant to any party's claim or defense and proportional to the needs of the case, litigants may potentially be entitled to compel production of information and communications generated or processed by AI platforms related to the facts in dispute. Fed. R. Civ. P. 26(b)(1); In re OpenAI, Inc., Copyright Infringement Litig., No. 23-CV-08292, 2025 WL 1652110, at *2 (S.D.N.Y. May 30, 2025). Just as local news headlines are replete with instances of internet searches as evidence in criminal cases[1], real-time AI “interactions” may likely be subject to the same disclosure requirements in civil litigation.
Similar to how email communications, instant messages, and internet search histories were once at the frontier of discoverable information, AI interactions, often resembling a combination of traditional search engine use and “correspondence” with a computer program that uses language models to generate human-like responses, are emerging as a new potential supply of information and documents that users previously may have believed was confidential. Earlier this year, a federal court in New York directed OpenAI, the creator and developer of ChatGPT, to preserve and segregate all output log data that would otherwise be deleted for potential disclosure to the litigants. In re OpenAI, Inc., Copyright Infringement Litig., 2025 WL 1442678, at *1 (S.D.N.Y. May 13, 2025), reconsideration denied, 2025 WL 1443780 (S.D.N.Y. May 16, 2025). Just as courts have evolved to recognize internet search histories and emails as potentially discoverable, there is an emerging risk that AI inputs and responses, mimicking correspondence, may be used to show how a litigant analyzed facts and information and created documents outside of the protection of attorney-client privilege or attorney work product protections.
While prompts crafted by legal counsel that contain an attorney’s mental impressions and opinions about how to utilize an AI tool have been found by some courts to consist of protectable work product(see Tremblay v. OpenAI, Inc., No. 23-CV-03223-AMO, 2024 WL 3748003, at *2 (N.D. Cal. Aug. 8, 2024); Concord Music Grp., Inc. v. Anthropic PBC, No. 24-CV-03811-EKL (SVK), 2025 WL 1482734, at *2 (N.D. Cal. May 23, 2025) (denying requests for production of AI search prompts but recognizing possibility of work product waiver where outputs were specifically relied upon and used in court pleadings), inputting privileged information into a non-confidential AI platform may result in waiver of protections from disclosure, as the terms of service of many AI tools grant providers rights to use or review user inputs. Parties seeking disclosure may argue that such disclosure to an AI vendor constitutes disclosure to a third party, waiving privilege, trade secret, or other confidential treatment. Of note, OpenAI CEO Sam Altman recently acknowledged in an interview that ChatGPT is not a privileged communication, and may be subject to disclosure.[2]
As this topic continues to develop, employers and litigants should prepare for AI interactions to be treated like other discoverable communications and establish policies concerning acceptable use, retention, and preservation of AI searches and chats, prohibit input of sensitive, confidential information or privileged communications into public AI tools, and ensure that potentially relevant AI interactions are subject to litigation holds in the event a dispute arises. Given the potential discoverability of AI prompts and outputs, clients should be advised that once litigation is reasonably anticipated, a duty to preserve potentially relevant evidence in their custody or control attaches, which may now extend to potentially relevant AI interactions, AI-crafted communications, or logs depicting objective facts input for eliciting AI-generated guidance and analysis. Taking this progression to the next logical step, failing to preserve potentially relevant AI interactions and data may subject litigants to allegations of spoliation.
AI searches sit at the intersection of research, communication, and work product. They are novel in form but not in principle: like emails, texts, or search histories, they may reveal facts, intentions, or thought processes central to litigation. As courts and litigants begin to push the boundaries of what can be collected and used as evidence, the safest course is to assume that AI searches and chats are discoverable, and act accordingly. As businesses and professionals continue learning how AI tools can be used to improve performance and efficiency, there must be policies and practices in place to ensure these uses and the corresponding factual inputs and outputs can be responsibly preserved, collected, and disclosed in litigation.
ENDNOTES
[1] https://turnto10.com/news/karen-read-murder-trial/story/testimony-continues-in-the-karen-read-murder-trial-after-the-jury-hears-angry-voicemails-boston-police-office-john-okeefe-may-7-2025; https://www.cnn.com/2023/01/18/us/brian-walshe-ana-walshe-google-searches
[2] https://www.forbes.com/sites/jasonsnyder/2025/07/27/openai-chatgpt-wants-legal-rights-you-need-the-right-to-be-forgotten/
[View source.]
